Examination of Doctor J. Thomas Brenna, Cornell.
Published on instrumentation and data processing of overlapping peaks and baselines. Specific application is another area, not steroids (fatty acid lipid proteins).
Q: Have you done IRMS with steroids?
A: Yes, have a USADA grant to create steroid standards, and to refine methods to permit more rapid analysis.
Q: Ever testified before?
A: Only small claims court. I'm an expert on my back yard.
Q: Other than doping control, what areas use IRMS?
A: Developed over the years for bio-geochemistry. Used extensively in some areas of research, water turnover, energy expenditure; testing breath for some metabolic issues for some health related diagnostics; ecology and environmental science.
Q: Part of your daily job to review IRMS chromatograms?
Q: Mr Suh mixed chromatograms from T/E and IRMS. Does the T/E ratio have anything to do with the IRMS result?
A: I'd have to say no.
Q: Would you use T/E to validate IRMS results?
A: Probably would not.
Q: What docs have you reviewed in this case?
A: Focused on IRMS results, and the Fraction 3's, and the results of the reprocessing on 4-5 May in Paris.
Q: Were you physically present for the reprocessing?
Q: Did you review the other B sample tests?
A: Yes, focused on the specific peaks associated with the steroid analysis. Looked at the peaks associated with doping analysis on all the chromatograms for peak integration.
Q: What is F3?
A: One that has 5b and 5a pdiols that are at issue.
Q: Do you have an opinion on whether the A and B are reliable.
A: Yes, I believe they are reliable.
A: First, I looked at the controls. They run several. . This lets me believe the instrument is right. Then they run the blank urine. These are processed the same as the athlete samples, and run at the same time as the athlete samples.
Q: Was the IRMS instrument working properly?
Q: What is the blank?
A: A big batch of urine, I assume, I didn't audit that, that they run before and after the athlete sample. One expects the interferences in the baseline will be similar between the blank and the athlete samples.
Q: did the blank values look good.
A: looked and verified that, but had trouble finding the page.
Q: What is the mix-cal IRMS?
A: To establish the ratios have good precision. They analyize it 3 times to check mean and standard deviation.
Q: Look at exhibit
A: They're OK.
Q: What about the mix-cal?
A: They look fine.
Q: How would you compare the number of controls LNDD ran to what you would normally see?
A: In our research lab, not an accredited lab, we run a number of controls, probably not as many as they run. I was prepared to see fewer; I was very impressed with what they did; they appeared to have controls at every level you'd want to control; they were redundant as well.
Q: If there would have been an operating pressure problem, would that have been reflected in the controls?
A: The way one establishes pressure is to run the controls, particularly the mix-cal controls, not the urine, to establish the machine has accuracy and precision in the desired range. Were the they to be outside the range, pressure is one of many parameters you'd look at to troubleshoot the problem. The cal results would indicate a pressure problem, and they did not.
Q: Were all the cal results in the range you'd expect for a properly operating instrument?
Q: Maintenance? Same question, would controls have revealed?
A: Yes, if the controls are right, the instrument is running properly.
Q: Let's talk about linearity. What is it?
A: It's the property by which the isotope ratio is identical within experimental error, and the amount of material introduced into the machine. You get a straight line.
Q: Have you reviewed LNDD's linearity data?
Q: Was it acceptable?
Q: Would any variations have made any difference in the delta-delta values reported?
[He doesn't ask or answer about the T/E linearity, only the IRMS...]
Q: Is error the same as measurement uncertainty?
Q: LNDD's single measurement for a single metabolite is 0.5 and 0.8 for delta-delta. Is that what you expect?
A: Yes, that is consistent with what I expect.
Q: You talked about mix-cal being derivitized like the athlete's sample. What does that mean?
A: The natural stuff can be hard to analyze; some places do a reaction to make the GC easier to do.
Q: What is the correction about?
A: The acetate came out of a bottle, when it's added to the sample, it shows up in the carbons in the GC. When we analyze the results, the number reflects the entire molecule. Then we do it to a known sample, and subtract the difference from the unknown.
Q: Did LNDD do that?
A: They did it. They applied a calculation to do the correction. I reviewed how they get the number they used, around -53/mil. They did it the right way, and I looked at how sensitive the result was to that correction, and it made no difference to the final results. Calculating delta delta is a robust way to make a measurement. Any source of systematic uncertainties get corrected by the subtraction. It's takes extreme values to affect the result.
Q: The resondents complained about separation and symmetry. Does it matter?
A: It was not great, but ok. Symmetry doesn't really matter to IRMS for well separated peaks, which is what I worry about.
Q: Look at 171 and 173.
A: 171 is the GC and 173 is the IRMS. They have the same peaks, and same retention times to tell us which is which.
Q: Peak separation for 5a and pdiol. pdiol looks well separated.
A: Yes, we talked about these. The small peak is quite small, which I'll return to.
Q: How do you know whether there is good separation for the 5a in the GCMS or IRMS?
A: For GCMS, there's another reference, for the masses, so it's possible to resolve, which you can't in the IRMS. This looks good for GCMS based on my experience.
Q: Now 173, the IRMS.
A: This is going to be easier. This was already addressed. We were looking here, and are they cleanly down into the baseline. There are always other additional molecules in the instrument. Which we'll explain how we correct for that. Let's look at the graphic for 4-5 May reprocessing.
Q: What are we looking at. (from GDC00970)
A: This is F3 for S17. Chose this one because it could be expanded easily.
Q: Does this come straight out of the electronic data files?
A: Yes, no corrections applied.
Q: What is the figure above?
A: Often we plot the instantaneous ratio, not included in the original doc pack, but is here.
Reveals a curious fact that molecules that have a C13 emerge from GC a little faster than the C12's, so you get overlapping peaks of C13s and C12s. So here we get C13 running into C12.
Q: What does this mean?
A: Criteria I use to establish separation. Look at this blowup of this region. We can argue if these are are baseline separated. When there is little signal from the GC, we get a noisy ratio. When a peak comes through, you get positive negative, baseline. What I see in the 5a is noise, band, baseline, noise, band, noise; this looks good to me. That peak is there if you no to look for it, but it is tiny.
Q: Would that peak have affected the 5a result?
A: No. Might have had a trivial affect on the 5b result. What I see here is that the result would be within experimental error. But the 5a was the positive, not the 5b.
Q: Is this bad chromatography?
Q: While we have this up, there were a series of questions about sloping baselines. Is this sloping baseline that would change the delta-delta results.
A: Not in my experience.
Q: Please explain.
A: There is always carbon from various sources. We're used to that. The procedures we use to draw baselines underneath peaks are intended to remove other contributions. With respect to sloping baselines, this would be an example. The next slide... One of the properties of ratios of c13/c12. Ratio trace is solid, straight horizontal line, which is an illustration how the sloping baseline doesn't matter. At least in our experience. Doesn't believe it does for LNDD either.
Q: What about the peaks following the 5a? Would they have any influence on the 5a peak?
Q: What about 171? Would the 8 after the 5a affect the result?
A: No. It might for diagnostic purposes if we thought there was a problem.
Q: 173 -- You don't see them as clearly in the IRMS. Does that suggest anything wrong with the IRMS?
A: No. They detect things different ways, the sensitivity of IRMS is related to the carbon content - this is predictable; the sensitivty of GCMS is sensitive to the molecular structure, and harder to predict. We'll see peaks in GCMS we won't in IRMS. So that doesn't concern me.
Q: Do any of those 8 peaks bear on the reliability of the 5a results?
Q: May 4-5th reprocessing. Were you there?
Q: Supervised by Dr. Botre, the panel's expert?
A: yes. We did it three ways. We'll call one "manual reprocessing". An analyst sat down, brought a chromatogram on the machine, told the OS/2 software to identify the peaks, checked manually to establish they were properly assigned, according to the SOP criteria.
Q: Was that there to see?
A: Yes. We use terms here loosely. The machine subtracts the background. I saw each chromatogram come up once, saw the QC step of adjusting the peak and most were adjusting. Then the ratio was computed. Didn't do it more than once. Watching what they were doing, over their shoulder, and thought what they were doing was mechanical, and identical from run to run to run.
Q: Let's look at a ratio trace...
A: You can see some vertical lines that are the limits of integration. They decide if each one goes one way or the other. Then they look at the ratio trace and match. They did it the same way every time. The did this every time, and I consider this a normal QC step. This uses feedback of the "instantaneous ratio", stopping when you get back to the background ratio.
Q: You call it QC work. Is this normal?
A: I insist on it, I believe it is common. You have to do it sometime, or you do it some othertime, like now. So it's routine.
Q: Last page of powerpoint.
A: This is SOP of their peak start/stop process. This shows right and wrong way to do it. This illustrates peak start and stop spanning the range of the peak, from baseline ratio to baseline ratio. This is an example where the limit of integration is set to far into the peak, showing software guessing wrong. It appeared to me the technicians would then move it correctly.
Q: So they were following the SOP?
Q: Third page of your powerpoint, excerpt from respondents supplemental brief. Where it says, when says it used manual instead of automatic? Is that correct?
A: No, that's not a right statement?
Q: Is that changing data, or normal processing?
A: This procedure is part of SOP; one expects uncertainty with the manual QC step will be included in their uncertainty estimates. It's what they normally do.
Q: As it says, did they achieve entirely different results? Is that true.
A: No. The numbers are not identical, but what we care about is within experimental error. Certainly the delta-deltas were the same, they came out as positives. So it's not a meaningful statement to me.
Q: The next comment says, they used a completely manual approach, is that true?
A: That's not what I described and saw. There were brief discussions before we went to do completely manual, as in taking the data into excel and adding numbers, but we didn't do that.
Q: When Cynthia for the A or Claire for the B, would they know when they set the baseline would they know how that would affect the delta-delta?
A: No, I don't believe so. They did their mechanical step, just once.
Q: Mr Suh said they did it 20 different times, does that make sense.
A: I didn't get that, I saw it just once.
Q: It was also done with no background subtraction, and then auto only. Whose idea was that?
A: Dr. Davis, the Landis expert suggested it.
Q: Did that make any sense to you?
A: No. There's background carbon going into the machine, and electronic offsets. Leaving background in was a bad idea. My intuition on that was wrong, because it seemed to have no particular result. Then the completely automated analysis, and seemed to get the same results for the delta deltas.
Q: As I've looked through the reprocessing data, I've noticed while doing the manual marking, the differences seem to be within error. But there seem to be larger differences in the other compounds. Would any of those changes affect the 5a measure?
Q: Did any of those other measures cause you to doubt the 5a result?
Q: The data was also reprocessed on Masslynx.
September 07: Hearing Award
October 07: Hue's Hearing Appraisal
November 07: Major document Release
January 08: Larry's Curb Your Anticipation
Monday, May 14, 2007
Examination of Doctor J. Thomas Brenna, Cornell.